EP3635680A1

EP3635680A1 - Computer-implemented method and device for automatically generating characterized image data and analysis device for inspecting a component

Info

Publication number: EP3635680A1
Application number: EP18753089.4A
Authority: EP
Inventors: Felix Buggenthin; Ulli Waltinger
Original assignee: Siemens AG
Current assignee: Siemens AG
Priority date: 2017-07-28
Filing date: 2018-07-27
Publication date: 2020-04-15
Also published as: DE102017213060A1; US11501030B2; WO2019020791A1; US20200167510A1

Abstract

The present invention relates to the automatic generation of characterized image data. For this purpose, a visual representation of a component is computed on the basis of existent structure data for the component, wherein the surface properties of the visual representation of the component can be varied based on predefined characteristic properties. Since, in the computation of such visual representations, both the component itself and the underlying characteristic surface properties are known, this information can be used for characterizing the corresponding parts in the visual representation in order to achieve automatic characterization of the computed visual representation.

Description

description

A computer implemented method and apparatus for automatic ^¬ tables generate marked image data and analysis of a component for checking sevorrichtung

The present invention relates to a computer-implemented method and to a device for automatically generating tagged image data of components, and to an analysis device for checking a component.

State of the art

The increasing digitization is currently leading to drastic changes, especially in the industrial environment. Open operating system solutions already exist that enable the connection between machines and physical infrastructures to the digital world. One such example is Siemens MindSphere, a cloud-based Internet of Things operating system that provides a platform to analyze industrial data, such as using artificial intelligence to implement many breakthrough applications. As efficiently defective components can be identified and then all the necessary steps for a repair be initiated turprozess example by means of opti ^¬ rule analysis.

The data processing in this case requires just at applicatio ^¬ gen artificial intelligence a comprehensive database. For training purposes, large amounts of manually acquired image data have hitherto been required for this purpose. In order to train an artificial intelligence system with these image data, however, these real image data must be marked accordingly. During this labeling process, relevant areas must be identified in the image data and each area must then be assigned one or more properties. For example, of a component, such as a mechanical shaft or a turbine blade, several photographs are created. In particular, a component can be photographed, for example, of several un ^¬ terschiedlichen perspectives. In addition, a component or possibly also several identical or similar components in different operating states, aging stages or with various types of damage can be recorded photographically. Then each of these photographs must be manually evaluated to identify the component in the corresponding image and, if necessary the component and characteristics like aging state possible damage, etc. zuzuwei ^¬ sen.

As part of the complexity of modern industrial plants and the resulting huge amount of data, it is already no longer possible to manually perform the capture of the images and the required labeling of the acquired image data. Even semi-automatic to ^¬ rates that support a user in specifying image data come because of the huge amount of image data and the complexity of dargestell ^¬ th in these image data systems very quickly reach their limits. In addition, the above-described acquisition of image data and the downstream marking only make it possible to capture image data from existing real systems and the damage or signs of wear that have occurred in the process. Furthermore, it is also necessary for the detection of image data of the real components to expose the corresponding components in each case so far that the components can be photographed. Consequently, should the wear of a component over the operating time to be photographically documented, in re ^¬ regular intervals the corresponding component must be removed, are fo ^¬ tografiert and then installed again. Again, this requires in complex technical systems a manageable effort.

There is therefore a need for automatic generation of tagged image data. In particular, there is a need for generating tagged image data called

Basis for data processing in complex technical systems can be used. In addition, there is a need to analyze components based on such image data.

Disclosure of the invention

The present invention discloses a computer-implemented method for generating tagged image data with the features of patent claim 1, an apparatus for automatically generating tagged image data with the features of patent claim 9 and an analysis device for inspecting a component with the features of patent claim 10 ,

Accordingly, it is provided:

A computer-implemented method for generating tagged image data of components. The method includes the steps of providing design data for a component and providing specifications for possible surface properties of the component. Furthermore, the method comprises a step for calculating a visual representation of the component. The visual representation is calculated using the provided design data and the provided specifications for the surface properties of the component. Finally, the method includes a step of automatically assigning predetermined labels to the computed visual representation of the component.

Furthermore, a device is provided for automatically generating tagged image data of components. The device comprises a first database, a second database and a Bildda ^¬ tengenerator. The first database is designed to To provide design data for a component. The second database is designed to provide specifications for possible surface properties of the component. The image data generator is designed to calculate a visual representation of the component. Specifically, the visual representation is calculated using the design data provided by the first database and the surface potential specifications provided by the second database. Furthermore, the image data generator is designed to automatically assign predetermined markings to the calculated visual representation of the component.

Finally, it is planned:

An analysis device for checking a component with a device according to the invention for generating marked image data, a reference image data memory and an evaluation device. The reference image data memory is configured to store the tagged image data generated by the apparatus for automatically generating tagged image data. The evaluation device is designed to receive a camera image of a component to be checked and to compare the received camera image with the marked visual representations stored in the reference image data memory.

ADVANTAGES OF THE INVENTION The present invention is based on the finding that the diagnosis of complex industrial plants of a database with a large number of identified image data is of great importance. Further, the present invention is based on the recognition that the Erstel- len such a data base for complex industrial An lay ^¬ and particularly identifying individual elements in the image data represents a major challenge to the con- can not be realized in a conventional manner, and in particular even with the use of human labor.

The present invention is therefore based on the idea to provide a solution with which it is completely automatically possible to generate labeled image data for components of industrial plants and to prepare them for further use in a digital environment. For this purpose, the present invention initially provides, not to obtain the image data of components in industrial plants in a conventional manner by photographing real components, but to calculate this image data digitally. Since components of modern systems are usually already constructed in a digital environment anyway, digital design data is also available for such components. Based on this digital design data, a three-dimensional model of the component can be easily calculated. Ins ^¬ particular it is possible to calculate a model of the component based on the existing design data. For example, convention ionel ^¬ le CAD data or any other design data can this be used as design data.

As a rule, the design data also provide information about the materials used in the component for which a visual representation is to be calculated. In addition, an arbitrary modification Impress ^¬ trächtigung, aging, damage, etc. may already be simulated and included in the calculation of visual representation in the calculation of the visual representation of the component. For example, by specifying various surface properties for the component, different different states of the component can also be simulated. For example, can be defi ned ^¬ which characterize the visual representation of the component in a predetermined state respectively different properties of a surface for various Alterungsstu- fen. In this way, for example, a corrosion example ^¬ rusting or the like, an increasing roughening or abrading a portion of the component, the Bil ^¬ extension of cracks, especially of micro-cracks or the like, stretching, compressing or bending as a portion of the Component, a change due to heat or any other change in the surface properties of the component simulated and included in the generated visual representation. This allows for loading Sonder efficient, multi-layered and detailed Ge ^¬ nerierung visual representations of the part.

Since the construction of each ^¬ the underlying conditions, particularly the respective specifications of the surface properties are known for each calculated visual display part, these known properties may also be assigned directly to the respective calculated visual representation. In particular, it is also known which image area in the visual representation corresponds to which part of the component and which surface properties of the respective calculation are based. As a result, a corresponding identification with the underlying parameters can be very precisely assigned to each individual part of a component in the visual representation.

As a result, image data of components can be produced while identifiers are assigned in these image data and individual areas meeting the underlying specifications for generating the image data very quickly, kostengüns ^¬ tig and efficiently in this way.

In this way it is also possible without any danger to obtain image data of components which correspond to a component with a particularly high and optionally critical stress. In particular, it is also possible to calculate image data which correspond to a component after a stress in a dangerous operating state, without the building part in such a dangerous operating condition to Operator Op ^¬ ben. Furthermore, image data can be obtained from perspectives in this way, which are not possible under real conditions or only with great effort.

In sum a fast, kostengünsti ^¬ ge, flexible and reliable generation of image data of components with simultaneous labeling of the image data is possible in this way.

In the previous and subsequent embodiments, a distinction is made in particular between the first database for storing the design data and the second database for storing the marked image data. Even if this different databases or different storage devices / storage media can be used in principle, it is also well mög ^¬ Lich, all data in a common database, in particular a common storage device / store a common storage medium and provide.

When referred to herein as second database storage unit can be, for example, in a knowledge base (ger .: Knowledge Base), which can be ge ^¬ uses for other applications, such as an automatic diagnosis of a component.

In one embodiment, the provided surface properties specifications include characteristic textures, particularly for predetermined areas of a surface of the component. For example, such characteristic textures may correspond to various stages of corrosion, such as rusting or other oxidation. Furthermore, different textures may also include different gradations of scratching a particular area of the component, different stages of abrading or grinding on a component, different stages of formation of cracks, in particular, for example, microcracks or the like, various stages of thermal action of heat or cold on the component, a deformation of a Teilberei ^¬ Ches on the component, in particular bending, stretching, upsetting or the like, a wave formation at a portion of the surface and any other physical or chemical action an area of the component. In this way, numerous different impairments, such as wear, aging, overuse or the like, can be simulated simply, inexpensively and also safely on the component in order to obtain a suitable optical representation of the component from it. In this case, the respective underlying specifications and the associated textures can be assigned automatically to the corresponding area in the generated visual representation in order to obtain a corresponding identification in the acquired image data.

According to one embodiment, the characteristic textures particular properties such as wear, aging, a predetermined load, a pollution or a potential failure probability of the component speci fied ^¬ can. This information can be taken into account in the automatic assignment of the given identifier to serve as a reference for the diagnosis of image data of real components.

According to one embodiment, the visual representation of the component is calculated from a given perspective and / or along a predetermined observation path. In particular, as a movement path, for example, a flight of a virtual drone along a flight path for the observation of a component can be specified. This enables a particularly flexible, universal generation of image data from almost any angle and perspective.

In one embodiment, the design data underlying the computed visual representation includes CAD data or manufacturing data for manufacturing the component. The- Such design data are already available for almost all components due to the modern production technology. Therefore, the visual representations of the component can be particularly easily calculated without further, possibly complex intermediate steps. The computation of the visual representations of the component can be generated, for example, on the basis of algorithms and / or graphics units such as those already used, for example, for calculating images in computer games or other optical simulations. In particular, a three-dimensional model can be calculated from the design data at first, on the basis of the visual representation of the component is generated ^¬ riert. The visual representation may correspond to a zweidi ^¬ dimensional image.

According to one embodiment, the automatic assignment of the predetermined markings to the calculated visual representations of the component comprises a multi-level hierarchical characterization, in particular of at least a part of the component. Through a multi-stage hierarchical characterization ^¬ tion, the individual properties of the component in the calculated visual representations can be very efficiently and accurately specified. According to one embodiment, the automatic assignment of the predetermined identifications to the computed visual representations of the component comprises at least one diagnosis for a predetermined error, an indication of a possible damage, a prognosis for a probability of failure and / or a repair recommendation. In addition, the predetermined indicia may also include any other parameters, properties, or values suitable for characterizing features of a portion of the component in the visual representation.

According to one embodiment, the computer-implemented method for generating the image data comprises a step of storing the visual representation of the component together with the assigned identifiers in a database, and a step of comparing an image of a camera to diagnosti ^¬ ornamental component with the stored visual display. In this way, by matching a camera image from a real component with the generated image data, the component can be characterized in a simple manner and, in particular, a statement about the current state of the component can be made from the assigned identifications. In this case, the comparison of the calculated and marked image data with real image data can not only comprise a direct pairwise comparison of image data. Rather, a more complex analysis of the image data may be performed based on the calculated tagged image data. For this purpose, methods based on artificial intelligence and neural networks or the like can be used.

According to one embodiment, the evaluation device of the analysis device for checking a component is designed to identify a match between at least a part of the camera image and a part of a calculated visual representation of a component. In this case, the evaluation device can be designed to output markings that correspond in the visual representation of the identified part of the component in the marked image data. Such markings can be used to diagnose a component to be tested in order to establish a rapid, reliable and efficient diagnosis of the component. Based on such a diagnosis a ra ^¬ specific, timely and cost-effective maintenance and repair of the component can be executed.

According to one embodiment, the analysis device comprises a control device. The control device can be designed to generate one or more control commands depending on the result of the comparison between the camera image and the marked image data. The generated control commands can then be output. These control commands may be, for example, a Shutdown command act to shut down a system with the corresponding component in a dangerous situation. In addition, the control command can also include a command to transfer a system into a predetermined operating state, which, based on the analysis of the component, enables a safe continued operation, at least for a predetermined period of time. In addition, the control command may also include a repair recommendation for the maintenance of the component to be diagnosed. Furthermore, the control command can also be used to cause an automatic order of required spare parts for maintenance or repair of the component. Of course, any additional control commands are possible on the basis of previous Ana ^¬ analysis.

According to one embodiment, the analysis device comprises an interface. This interface can be designed to receive at least one camera image of the component to be checked. This interface can play as acting accession to a wired interface which is coupled directly or via a data bus or the like with ei ^¬ ner or more cameras. In addition, it can also be a radio interface which receives image data from one or more cameras via a wireless transmission. For example, there is also a reception from

Image data from a mobile phone, a tablet computer or any other mobile or portable system with an image capture device possible. According to one embodiment of the analysis device, the evaluation device of the analysis device is designed to carry out the comparison between the camera image and the stored marked visual representations based on artificial intelligence. In particular, a so-called Convolutional Neural Network (CNN) can be implemented in the analysis device. In a CNN, for example about "folding neural network", it is a

feedforward artificial neural network. This is a concept inspired by biological processes in the field of machine learning.

The above embodiments and developments can, as far as appropriate, combine arbitrarily. Further refinements, further developments and implementations of the present invention also include combinations of feature features of the invention which have not been explicitly mentioned above or described below with respect to the exemplary embodiments. In particular, the person skilled in the art will also add individual aspects as improvements or supplements to the respective basic forms of the present invention.

Brief description of the drawings

The present invention will be explained in more detail with reference to the exemplary embodiments indicated in the schematic figures of the drawings. Showing:

1 shows a schematic representation of an analysis device for checking a component with a device for the automatic generation of image data according to an embodiment;

Fig. 2: a schematic representation of marked

Image data as underlying an embodiment; and

3 is a flowchart underlying a computer-implemented method for generating tagged image data according to one embodiment.

Description of embodiments

1 shows a schematic representation of an analysis device for checking a component according to an embodiment. The analysis device may, for example, a Apparatus 10 for automatically generating tagged image data. This apparatus 10 for automatically generating tagged image data comprises a first database 11 in which design data for one or more components is provided. In these

Design data may be, for example, CAD data or any other design data for a part. Since the construction of components in recent years increasingly digital systems are used, the design data of these components are usually already in digital form. This design data of the component can be used as a basis for the generation of ge ^¬ marked image data. For this purpose, the corresponding design data in the first database 11 can be stored and provided. Furthermore, specifications for possible surface properties of components can be provided in a second database 12. These specifications of surface properties can characterize different material properties of a component. For example, in the second database 12

Textures for different material states or surface states of a component can be specified. For example, textures for different gradations ei ^¬ ner corrosion specified on a material surface advertising to. For example, different degrees of gradation of the

Rostens an iron surface and the corresponding textures of the respective surface are specified. Further, wear by abrading a material surface and the textures corresponding thereto in the second database 12 may also be specified. Also, other characteristics such as forming grooves or grooves due to rubbing of two material surfaces on each other can be specified. Further, thermal or thermal exposure to a material may also result in a characteristic optical change which may also be specified in the second database 12. For example, contact by strong heat ^¬ exposure to a metal called tarnish on, from which shows the maximum temperature effect. Corresponding colors or textures can be specified in the second database 12 for the corresponding surfaces of a component. Further characteristic features specified in the second database 12 are, for example, soiling, in particular also soiling, which may be the result of improper use of the component, the formation of cracks, in particular microcracks, during normal operation or overstressing of the component, characteristic deformations,

Strains or compressions of a part of the component and associated changes in the surface of a part of the ^¬ construction part, but also changes in material thickness and associated changes in the visual appearance of the component and any other characteristic visually perceptible changes to the surface of the component.

Furthermore, for example, 13 positions can be specified in a third database, from the perspective of which a visual representation of a component is to be calculated. Here, individual positions can for example be specified, a visual representation be from their perspective ^¬ expects to be, or is it more complex motion ^¬ paths, such as lines, curves, polygons or similarity ^¬ royal possible along which a virtual camera be ^¬ moves. In this way, for example, virtual flights of a drone or the like can be simulated, which moves along a component or around a component. In particular ^¬ sondere can be defined in this way also prospects which are not or only with great effort possible to a component when installed under real conditions.

Even though the first database 11, the second database 12 and the third database 13 are described herein as a single, separate databanks ^¬ ken, it is also possible to store and provide the data of two or all three databases in a gene insamen storage means. In advance, as in the following, a component is considered in each case. The term "a component" is not to be understood as a single individual component. Much more can be understood as a component and complex machine ^{arrangements of} any number of individual parts. Thus, the visual representation of the component to be generated can also include the representation of a complete machine or a subassembly of a machine. In the case of a power plant, for example, as a component, a complete turbine, a drive shaft including bearings or any other composite of several machine parts or components can be understood.

On the basis of the information provided in the databases 11 to 13, an image data generator 20 can calculate a visual representation of the component. For this, a dreidimensiona ^¬ les model of the component can be calculated, for example, from the design data first. In a further step, the surface of this three-dimensional model can be determined according to the specifications for the surface properties. In this way it is possible, for example, to include certain stresses, damage, wear or the like in the calculation of the visual representation of the component. Based on the underlying data, the component can be simulated in almost any state and a visual representation based on it can be generated. In particular, visual representations of the component are possible, which are due to over ^¬ claims, possibly in dangerous Betriebszu- states. Thus, such image data can be obtained without danger, without the component must be placed tatsäch ^¬ Lich in such a dangerous or at least critical operating condition. Since the information underlying the calculation of the image data is completely known, a corresponding parameterization can also be assigned to the individual regions in the image data. For example, on a predetermined area of the component calculates an optical representation, which is on a high heat ^{zurückzufüh ¬} ren, the corresponding area in the calculated visual representation can be characterized accordingly, ie

be parameterized. For example, this portion of the component, the designation of the part, the name of the corresponding portion in the component, the use ^¬ te material in this portion of the component, and the further underlying the effect of heat are assigned in this partial area of the corresponding partial area of the image. As can already be seen from this previous example, the assignment of markings to subareas of the calculated visual representation can be based on a multi-level hierarchical structure. In principle, any other types of assignment of labels to subregions in the image data are also possible. In principle, a common characteristic or marking can also be assigned to a plurality of subregions of the component in the visual representation.

After image data with a visual representation of a component using the data from the databases have been calculated 11 to 13 and these have been provided with suitable Distinguishing ^¬ voltages, the generated image data with the identifiers in a reference image data memory 30 can be stored. The storage of the image data, in particular the assignment of the marking to subregions in the image data, can take place in any structure. In particular, any databases with a 1: n structure or even an m: n structure can be used to assign labels to the image data.

The tagged image data stored in the reference image data memory 30 may be used, for example, to train an analysis system 40. In this

Analysis system 40 may be any system which, based on provided marked image data, an evaluation, in particular an evaluation from image data of real components. Such a system may be, for example, an artificial intelligence system. For example, neural networks may also be trained based on the provided computed image data to enable analysis of real device image data.

For checking a component, the evaluation device 40 can be provided with a real image of a component. This real image of the component can be recorded, for example, by means of a camera 50. The camera 50 can be coupled with ^¬ means of a wired connection to the evaluation device 40th In addition, cable-free communication ^paths between the camera 50 and the evaluation device 40 are also possible. For example, in a machine shop, one or optionally also more cameras 50 may be installed that detects a component or a system with egg ^¬ NEM or more components visually. The image data acquired thereby can be provided to the evaluation device 40, which then makes a comparison or a comparison

Performing evaluation based on the marked image data stored in the reference image data memory 30. In this way, monitoring and early fault detection can be performed continu ^¬ ously during operation of a system. In addition, image data can also be acquired from a component at predetermined intervals, for example during maintenance or the like, in particular when a system is at a standstill, and these data are provided to the evaluation device 40. Also in this case, the evaluation device 40 based on the marked

Image data in the reference image data memory 30 perform a diagnosis in order to make the most accurate possible statement about the aktuel ^¬ len state of the component, which has been optically detected by the camera 50.

The analysis can be, for example, a ^{comparison of} the acquired image data of the real component with the image data stored in the reference image data memory 30 act. In addition, any further gegebe ^¬ appropriate, more complex analyzes on basis of data stored in the Refe ^¬ ence image data memory 30 image data are possible. On the basis of the analysis in the evaluation device 40, the current state of the component to be checked can then be determined. In addition, a forecast of the further development of the component and in assess ^¬ Zung over the still expected remaining life of the component is optionally also possible.

If the component is damaged, such as cracks or the like, which detek- based on the data stored in the Re ference ^¬ image data memory 30 image data may be advantage, such a damage can be diagnosed. If necessary, then an order process for a required spare parts initi ^¬ ated can be automatic. Furthermore, further repair measures can be triggered in order to achieve a rapid repair of the component to be diagnosed. In this way, the availability of the component and the associated entire technical system can be increased.

FIG. 2 shows a schematic representation of automatically generated image data with corresponding identifications, as they are based on an embodiment. In the image shown in Fig. 2 is a highly simplified schematic representation. The representation of the building ^¬ part comprises a component A, which consists of two sub-elements Aa and Ab. The specification of the component can ^¬ on the design data it ^¬ follow when, for example, based here. On the basis of this design data, a three-dimensional model can first be calculated. The subelement Ab can furthermore be subdivided, for example, into two subparts Abi and Ab2. Furthermore, for example, on the part ^¬ element Abi a special stress can be assumed. This particular stress can, for example, lead to a change in the surface and a correspondingly modified result in the texture of the visual presentation. In Fig. 2 this is represented by marking Abl-x. Since the respective framework conditions are known in the gene ^¬ ration of the visual representation of the component A by the design data and the characteristic properties of the surface properties of the component, in each case a corresponding marking can be assigned in corresponding sub-areas of the visual representation. Such image data can then be stored, for example, in the previously described reference image data memory 30 and used as training or reference data for a subsequent analysis of similar components.

FIG. 3 shows a schematic representation of a sequence diagram on which a computer-implemented method for generating tagged image data of components is based. Step S1, design data for a part is provided. Furthermore, speci ^¬ fications for possible surface properties of a component are provided in step S2. In a further step Darue about ^¬ may optionally contain one or more items for possible perspectives for the image data to be calculated are specified addition. In step S3, then a calculation of a visuel ^¬ len representation of the component takes place. The component is calculated using the provided design data and the provided surface properties specifications. Furthermore, the calculation of the visual representation for the specified perspectives can be carried out. The specified perspectives may be one or more predefined points in the space or optionally also a trajectory of a virtual camera.

After the visual representation of the component has been calculated on the basis of the described boundary conditions, in step S4 an automatic assignment of pre-determined agreed markings of the calculated visual representation of the component. The automatic assignment of Distinguishing ^¬ voltage are measured on the basis of parameters that defines taken as a basis of loading ^¬ statement of visual representation in step S3. In addition, the embodiments that have already been described above in connection with the device also apply to the method for generating the marked image data. In summary, the present invention relates to an auto matic ^¬ generating marked image data. The marked image data can be used in modern industrial plants for the monitoring and diagnosis of components, plant components up to entire industrial plants. The marked image data serve as reference or training basis for the evaluation of image data of real plant components.

In particular, the present invention relates to automatic generation of tagged image data. For this purpose, a visual representation of the component is calculated on the basis of already existing design data for a component, wherein the surface properties of the visual representation of the component can be varied based on predetermined characteristic properties. Since the calculation of such visual representations both the component itself, and the underlying characteristic surface properties are known, this information can be used to identify the corresponding parts in the visual representation to allow automatic identification of the calculated visual representation. In this way, image data can be automatically generated and in the same without the need for a ma ^¬ Nuelles intervention is required. This makes it possible to produce marked image data on a large scale with a high quality.

Claims

claims

1. A computer-implemented method for generating ge ^¬ marked image data of components, comprising the steps of:

Providing (SI) design data for a component;

Providing (S2) specifications for possible surface properties of the component;

Calculating (S3) a visual representation of the component using the provided design data and the provided specifications for the surface properties of the component; and

Automatically assigning (S4) predetermined identifiers to the computed visual representation of the component.

2. The method of claim 1, wherein the specifications include characteristic textures provided for predetermined Be ^¬ rich the surface of the component.

3. The method according to claim 2, wherein the characteristic

Textures include properties such as wear, aging, stress, contamination and / or potential failure probability of the component.

4. The method according to any one of claims 1 to 3, wherein the

Step (S3) for calculating the visual representation of the component, the visual representation calculated from a given perspective and / or along a predetermined movement path.

5. The method according to any one of claims 1 to 4, wherein the design data include CAD data or manufacturing data for the production of the component.

6. The method according to any one of claims 1 to 5, wherein the au ^¬ matic allocating (S4) includes the predetermined markings on the calculated visual representation of the component a multi-level hierarchical characterization of at least a portion of the component.

7. The method according to any one of claims 1 to 6, wherein the au ^¬ tomatic assigning (S4) of the predetermined identifiers to the calculated visual representation of the component a diagnosis for a given error, a possible Beschädi ^¬ tion, a forecast for a probability of failure and / or includes a repair recommendation.

A method according to any one of claims 1 to 7, comprising a step of storing the visual representation of the component together with the assigned tag in a database, and a step of comparing a camera image of a component to be diagnosed with the stored visual representation.

9. An apparatus (10) for automatically generating tagged image data of components, comprising: a first database (11) configured to provide design data for a component; a second database (12) adapted to provide specifications for possible surface properties of the component; an image data generator (20) adapted to compute a visual representation of the device using the design data provided by the first database (11) and the surface potential specifications provided by the second database (12), and the computed visual representation automatically assign predetermined identifications to the component.

10. A component inspection apparatus comprising: an apparatus (10) for generating tagged image data according to claim 9, a reference image data memory (30) adapted to store said tagged visual representations; an evaluation device (40) which is adapted to compare a camera image of a component to be checked for emp ^¬ collected and the received camera image with the stored in the re ference ^¬ image data memory (30) marked visual representations.

11. The analyzer of claim 10, wherein the evaluator is configured to identify a match of at least a portion of the camera image with a portion of the visual representations and to output an identifier associated with the visual representation of the identified match is.

12. Analysis device according to claim 10 or 11, with a control device which is designed to generate control commands depending on the result of the comparison with the marked visual representations and to output the generated control commands.

13. Analysis device according to claim 10 to 12, wherein the evaluation device (40) is adapted to perform the comparison between the camera image and the stored labeled visual representations based on artificial intelligence, in particular based on a Convolutional Neural Network.

14. Control device according to one of claims 10 to 13, wherein the analysis device (40) comprises an interface, which is adapted to receive at least one camera image of the to-be tested ^¬ component.